Skip to content

Conversation

@teddybear082
Copy link

-add commandline option for gradio app.py, --no_vm, to run omniparser on user's local computer rather than requiring windows VM

-persist commandline option state for session across scripts by using environment variable

-add powershell script for easier install of models on windows

-modify gitignore to ignore virtual environment (venv) and bat files if any that may be added by the user , as well as temp directory for screenshot pictures

-add separate requirements.txt for gpu use

…ip VM

-command line argument --no_vm to skip check of VM

-check heartbeat of server that runs VM, and if no heartbeat, assume user is in no_vm mode and operate directly on computer (ideally this should be passed through from gradio to underlying tools but still need to improve this rather than just assuming)
-add weights_download.ps1 for users to run in windows terminal as alternative for downloading model weights

-add requirements_cuda for working version of cuda torch for nvidia gpus
-allow command line option of --no_vm to persist across various scripts to allow user to run on local machine without a vm by setting an environment variable for the session

-fix other bugs and typos
@teddybear082
Copy link
Author

@microsoft-github-policy-service agree

@microsoft-github-policy-service agree

@YangYa666
Copy link

Hello, I am a Chinese user. I can't access the api provider here, such as grop, but I have the deepseek api. I don't know how to change the code so that I can use the deepseek api without using grop

@teddybear082
Copy link
Author

Hello, I am a Chinese user. I can't access the api provider here, such as grop, but I have the deepseek api. I don't know how to change the code so that I can use the deepseek api without using grop

This is off topic for this pull request but if deepseek has an openai-compatible API with tools use, you might be able to get away with hacking in support by modifying this file: https://github.com/microsoft/OmniParser/blob/master/omnitool/gradio/agent/llm_utils/oaiclient.py in your own local install to hard code the base url as deep seek’s API URL instead of OpenAI’s, and hard code whatever model name deep seek requires you to use. Then in the user interface select openai as the provider (which is the default anyway) and put your deepseek API key in. It might work but there’s no guarantees. Proper built in support would require more steps.

@kaiseu
Copy link

kaiseu commented Mar 9, 2025

does the local machine support Mac or only windows?

@teddybear082
Copy link
Author

does the local machine support Mac or only windows?

I don't have a mac to test so I have no idea, it works on windows. However, all I'm doing is redirecting pyautogui commands to run directly on the local machine rather than searching for a virtual machine and sending the commands to it. According to pyautogui it works with Mac so unless something else in this library requires windows seems like it should work although if you encounter errors on mac maybe see if you need to install an additional dependency based on the pyautogui documentation: https://pypi.org/project/PyAutoGUI/.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants